an introduction to image segmentation and object-oriented...
TRANSCRIPT
An Introduction to Image Segmentation andObject-oriented Analysis
Wayne Walker and Ned Horning
University Mulawarman, Samarinda, Indonesia November 8-12, 2010
Images are made up of objects and not pixels!!
• Process of grouping pixels
• Intent is usually to simplify the image into meaningful pixel groupings (i.e., segments/objects)
• Segments are relatively homogeneous with regard to one or more characteristics
What is Segmentation?
• Bottom up• Image simplification• Image classification• Image compression• Edge detection• Object-Based Image
Analysis (OBIA)
• Top down• Feature extraction• Object recognition
Uses of Segmentation
• Classification is performed on objects instead of pixels (increased signal-to-noise ratio)
• Classification is performed using meaningful objects
• Algorithms operate on many more object-related features than typically available with pixel-based approaches
• Reduces salt and pepper effect of classifications
• Speeds up processing
Why Segment Images before Classification?
• Features in an image can vary from fine to coarse scale
• Need to find a balance (compromise) between too many and too few segments
• Multi-scale approach identifies features at appropriate scales Images from: http://fuentek.net/technologies/rhseg.htm
Segmentation and Scale
Region growing• Find similar pixels from
a seed and neighboring pixels
Watershed detection• Mostly for gray-scale
images• Treats image like a
topographic surfaceMean shift
• Used for segmentation and filtering
• Uses feature space and spatial domain
From: Mean shift: A robust approach toward feature space analysis
Algorithms for Segmenting Remotely Sensed Images
Spectral • Mean• Variance• Range• Ratios
Spatial• Area• Shape• Location• Context / Neighborhoods
Information Derived from Segments
• eCognition
• Spring
• RHSEG
• OTB/Monteverdi
Software for Image Segmentation
• Most popular segmentation software
• A stand-alone product for object-based image analysis
• Uses region growing
• eCognition now owned by Trimble
• www.ecognition.com
eCognition
eCognition Background: Software Features
• Multi-scale (hierarchical)
• Multi-source
• Multi-resolution
• Multi-temporal
eCognition: Multi-scale/Multi-source Segmentation
• Freeware GIS / Image processing
• Region growing and watershed segmentation
• http://www.dpi.inpe.br/spring/english/
Spring
• Open source software from NASA
• Stand-alone package for unsupervised image classification using sub and super-sets of segments
• User labels each segmentation level
RHSEG
• Open source software
• OTB is a software library and Monteverdi is the application
• Integrating with QGIS
• Watershed, region growing, level sets, and mean shift segmentation
OTM/Monteverdi
Segmentation software evaluations• http://www.ioer.de/segmentation-evaluation/
Berkeley segmentation dataset and benchmark• http://www.eecs.berkeley.edu/Research/Projects/CS/vis
ion/bsds/
Segmentation Resources
Thank you!